How Deeploy gives explainable AI (XAI) a central place in MLOps

As artificial intelligence (AI) gets weaved into every aspect of our lives and AI companies reach a stratospheric value, there is a critical AI problem that many are failing to discuss. According to Tim Kleinloog, co-founder and CTO of Deeploy, is that “it was and still is extremely hard to get AI into production successfully.”

While there seems to be a new AI model launching left, right and centre, there is a problem with those models that is less discussed. Kleinloog points out that there is a general problem with trust in production AI systems. In 2020, Kleinloog joined hands with Maarten Stolk, Bastiaan van de Rakt and Nick Jetten to co-found Deeploy as a startup that offers a technical solution to the roadblocks of trust in AI.

Next to Kleinloog, the leadership team includes Maarten Stolk (CEO), Sofia Karali (Head of Marketing), Markus Heid (Head of Operations), Robert Jan van Vugt (Head of Commerce), and Robbert van der Gugten (Head of Engineering). This dynamic Deeploy team is on a mission to give explainable AI (XAI) a central place in ML operations and here’s how they aim to make AI systems truly accountable.

Importance of XAI

Kleinloog highlights that explainable AI (XAI) is essential in human-in-the-loop (HITL) systems to provide decision-makers with the necessary context to make their AI-supported decision or correct the model. Unfortunately currently most AI systems are being designed without humans in the loop in mind, often blocking the final steps to production.

Take the example of an AI-supported doctor who needs to diagnose a patient based on an X-ray photo. Without XAI, the doctor can either decide to follow the AI’s advice or ignore it. With XAI, Kleinloog explains that doctors can compare their own expert knowledge with the explanation of the AI model. This process can improve decision-making and allow for detailed feedback from the doctor on the model’s decision.

In addition, XAI also increases the understanding and trust of a model developer, of the model behaviour, and intent. If you are wondering why there isn’t enough public discussion of XAI then it’s primarily that most companies recognise the importance of XAI but sometimes need a bit of convincing and even educating on the vitality of XAI.

“For us it is so obvious that you should not wait to make AI systems responsible,” says Kleinloog. He adds, “Especially when regulation requires transparency you see companies look at solutions based on XAI.”

Making XAI central in MLOps

How does Deeploy make XAI central in MLOps? Kleinloog explains that XAI should be present as part of model deployment and treated with the same importance as model development. He argues that only such an approach will create the possibility to interrogate the model on user request.

With the hype around GenAI reaching a fever pitch, Deeploy sees XAI truly becoming a central place in ML operations. “People are starting to recognise the risks of AI and realise that control and transparency are essential for its responsible use,” explains Kleinloog.

Deeploy is already serving millions of explanations per day for its customers. The explanations play the vital role of offering additional content needed to make AI-supported decisions in human-in-the-loop systems or correcting the model.

Ambitious Cloud play

Deeploy is more and more recognised as the enabler for high-impact AI use cases.. Kleinloog says the biggest cloud challenge regularly faced by Deeploy is integrating with the ever-changing AI ecosystem where all the major cloud providers are playing their part. He says scaling the cost efficiently is another major challenge with use of cloud-based solutions.

To overcome these challenges, Deeploy has found a trusted partner in DoiT, a Silicon Valley upstart providing cloud-driven organisations with intelligent technology and multi-cloud expertise to save both time and money. “DoiT has been a reliable knowledge partner in proposing a cost-efficient design solution,” says Kleinloog.

One of the ways Deeploy saves on time and money is by using DoiT as their main entry point for AWS support. The startup benefits from DoiT’s expertise and cost saving solutions Deeploy likes the “everybody wins” business model of DoiT, which helps the Dutch startup get better prices and support without spending additional money and “potentially saves money.”

As AI becomes the cornerstone of everything technology, Deeploy aims to offer the latest cloud technologies in a cost-efficient manner, with DoiT’s expertise providing a competitive edge.

Evolution of XAI

Deeploy envisions XAI having a breakthrough in the coming years. Kleinloog believes understanding the model internals is not only needed to keep control of current state-of-the-art models but also allows us to create the next generation of GenAI models. With regulations such as the EU AI Act coming into force, Kleinloog expects to see transparency efforts from companies like OpenAI to stay active in the European market.

Deeploy’s goal is to contribute to the alignment between humans and AI. “We believe that XAI already has a role in improving responsible use and trust in AI systems and that will only increase in the future,” Kleinloog adds.

Deeploy wants to make state-of-the-art XAI methods available for use in practice and it doesn’t want to do that alone. Backed by €3.5M in funding and DoiT’s multi-cloud expertise, the Utrecht-based startup is uniquely positioned to propel XAI to common parlance in the AI world.

DoiT helps digital native companies implement cost controls and optimisations in the cloud to drive business growth. Sign up for a free trial of the product to learn how DoiT’s intelligent software and unrivalled cloud expertise can save you time and money.